Towards Edge-Based Deep Learning in Industrial Internet of Things

Area of Science:

• Industrial Internet of Things (IIoT)
• Deep Learning
• Edge Computing

Background:

• Industrial Internet of Things (IIoT) systems connect devices for industrial monitoring and control.
• Deep learning in IIoT requires significant computation, typically cloud-based, leading to network congestion.
• Transmitting IIoT data to the cloud for deep learning impacts network performance.

Purpose of the Study:

• To propose an edge computing-based deep learning model for IIoT systems.
• To reduce data transmission demands and mitigate network congestion in IIoT networks.
• To optimize deep learning models for the limited computational power of edge nodes.

Main Methods:

• Leveraged the fog/edge computing paradigm.
• Developed an edge computing-based deep learning model.
• Designed a mechanism to optimize deep learning models for reduced computational requirements.
• Implemented a testbed in Google Cloud and deployed a Convolutional Neural Network (CNN) model.
• Evaluated the approach using a real-world IIoT dataset.

Main Results:

• The proposed model effectively reduces network traffic overhead in IIoT.
• Classification accuracy is maintained comparable to baseline schemes.
• The approach mitigates network congestion caused by data transmission.

Conclusions:

• Edge computing offers a viable solution for deep learning in IIoT.
• Optimized deep learning models can be deployed on edge nodes with limited resources.
• The proposed method enhances IIoT network performance and supports intelligent applications.